Improving collaborative filtering’s rating prediction accuracy by considering users’ dynamic rating variability

Users that populate ratings databases, follow different marking practices, in the sense that some are stricter, while others are more lenient. Similarly, users’ rating practices may also differ in rating variability, in the sense that some users may be entering ratings close to their mean, while other users may be entering more extreme ratings, close to the limits of the rating scale. While this aspect has been recently addressed through the computation and exploitation of an overall rating variability measure per user, the fact that user rating practices may vary along the user’s rating history time axis may render the use of the overall rating variability measure inappropriate for performing the rating prediction adjustment. In this work, we (1) propose an algorithm that considers two variability metrics per user, the global (overall) and the local one, with the latter representing the user’s variability at prediction time, (2) present alternative methods for computing a user’s local variability and (3) evaluate the performance of the proposed algorithm in terms of rating prediction quality and compare it against the state-of-the-art algorithm that employs a single variability metric in the rating prediction computation process.